The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local orchestration tools and an engine binary (referenced via the {{INVOKE}} placeholder) to manage development states, logs, and workflow transitions. It incorporates explicit logic for shell-safe argument escaping (supporting both POSIX and PowerShell quoting) when passing user-provided text to these commands.
[INDIRECT_PROMPT_INJECTION]: The orchestrator has a inherent attack surface as it processes data from external tool directives, project files, and user input.
Ingestion points: Ingests data from orchestration directives (e.g., narration, question, rules_content), user-supplied arguments, and project-local files like stage_file and consumes manifests.
Boundary markers: The instructions mandate verbatim reproduction of tool-generated narration and use structured question rendering (via AskUserQuestion) to maintain clear separation between system instructions and external content.
Capability inventory: The skill is capable of executing shell commands, reading project files, and dispatching tasks to sub-agents.
Sanitization: It implements shell-safe escaping for user-supplied strings to mitigate command injection risks at the interface with local tools.
[DYNAMIC_EXECUTION]: The skill dynamically dispatches work to specialized agent personas (e.g., aidlc-product-agent, aidlc-architect-agent) using a Task() mechanism based on the current workflow stage and instructions provided by the local orchestration engine. This behavior is standard for a multi-agent workflow system and is governed by a central conductor persona.

aidlc